To facilitate robust understanding of the processes and properties that govern a groundwater system, managers need data. However, this often requires them to make difficult decisions about what types of data to collect and where and when to collect it in the most cost - effective manner. This is where data worth analysis, which is based on predictive uncertainty analyses, can play an important role. The ‘worth’ of data is defined here as the reduction in uncertainty of a specific prediction of interest that is achieved as a result of a given data collection strategy. With the use of data worth analysis, the optimal data types, sample locations, and sampling frequencies can be determined for a specific prediction that informs, for example, management decisions. In this study a data worth method was used to optimize data collection when predicting pumping - induced stream depletion (water quantity section) and when predicting changing nitrate concentrations as a result of land-use change (water quality section). Specifically, the First Order Second Moment (FOSM) based data worth method was employed. This thesis also builds upon previous work which explores the impacts of spatial model parameterisation on the performance of the data worth analysis in the context of stream depletion assessments. A transient groundwater model was developed, using the MODFLOW-NWT software, and a steady state transport model was developed, using the MT3D-USGS software for the mid-Mataura catchment located in Southland, New Zealand. The ‘worth’ of both existing and additional potential monitoring data were investigated. In addition, and for only the water quantity part of the thesis, three spatial hydraulic parameter density scenarios were investigated to assess of parameter simplification on the performance of the data worth method: 1) distributed pilot-point parameters, 2) homogeneous parameters, and 3) grid-cell based parameters. The water quantity (stream depletion) predictions were made at 2 key locations: (i) the catchment outlet at Gore and (ii) the outlet of a spring-fedstream (McKellar Stream). The water quality prediction (change in nitrate concentration due to land-use change) was made at 7 locations 4 key surface water locations, 2 town supply bores at Gore and one additional groundwater location further upstream. For the water quantity predictions, results show that the existing transient groundwater level data resulted in the largest reduction in uncertainty for the predictions examined. Because the low flow predictions at Gore were integrating predictions, the most uncertainty reducing observations were scattered through the catchment area with a focus on the north-west. This coincides with the recharge zone (which means that there are large water level fluctuations and hence a larger ‘signal to noise’ content in the groundwater level data). In contrast, because McKellar Stream is a discrete prediction (in this case, because McKellar Stream is spring-fed), the observations directly surrounding the stream reduce the uncertainty the most significantly. The impact of parameter simplification in the water quantity modelling showed that the data worth analysis using the grid-cell based parameterisation were very similar to those using pilot-points. However, when using the homogeneous parameterisation, the data worth results became corrupted by the lack of spatial variability available in the parameterisation. Indicating that spatial heterogeneity is needed when predicting low flows, as was shown by previous studies. However, the computational time associated with performing data worth uncertainty analyses is much higher with a grid-cell based parameterisation. A pilot-point based scheme should perhaps therefore be considered a favourable option. For the water quality predictions, results showed a strong correlation between the hydraulic conductivity, porosity and denitrification. This is likely because the hydraulic conductivity and porosity provide information about the velocity of the groundwater for a given hydraulic - head gradient, which provides information about the amount of time available for denitrification to take place in the soil substrate. Next to that, results showed no distinct difference between surface water and groundwater predictions when predicting changing nitrate concentrations, but they showed that the spatial data worth patterns depended on the proximity of the prediction location to the denitrifying areas. Overall it can be concluded that spatial parameterisation is needed when performing a data worth study for stream depletion predictions, however a more detailed parameterisation than pilot – points does not provide significantly more information. Next to that, it can be concluded that the spatial data worth patterns when predicting low flows mainly depend on if the predictions are integrating or discrete predictions. Lastly, it can also be concluded that the data worth patterns when predicting change in nitrate concentration depend on the proximity of the prediction location to the denitrifying areas.
| Identifer | oai:union.ndltd.org:UPSALLA1/oai:DiVA.org:su-160851 |
| Date | January 2018 |
| Creators | op den Kelder, Antonia |
| Publisher | Stockholms universitet, Institutionen för naturgeografi |
| Source Sets | DiVA Archive at Upsalla University |
| Language | English |
| Detected Language | English |
| Type | Student thesis, info:eu-repo/semantics/bachelorThesis, text |
| Format | application/pdf |
| Rights | info:eu-repo/semantics/openAccess |
Page generated in 0.0019 seconds